ebook img

The Challenge of Anticipation: A Unifying Framework for the Analysis and Design of Artificial Cognitive Systems PDF

294 Pages·2008·3.95 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview The Challenge of Anticipation: A Unifying Framework for the Analysis and Design of Artificial Cognitive Systems

Lecture Notes in Artificial Intelligence 5225 EditedbyR.Goebel,J.Siekmann,andW.Wahlster Subseries of Lecture Notes in Computer Science Giovanni Pezzulo Martin V. Butz Cristiano Castelfranchi Rino Falcone (Eds.) The Challenge of Anticipation A Unifying Framework for theAnalysis and Design ofArtificial Cognitive Systems 1 3 SeriesEditors RandyGoebel,UniversityofAlberta,Edmonton,Canada JörgSiekmann,UniversityofSaarland,Saarbrücken,Germany WolfgangWahlster,DFKIandUniversityofSaarland,Saarbrücken,Germany VolumeEditors GiovanniPezzulo CristianoCastelfranchi RinoFalcone IstitutodiScienzeeTecnologiedellaCognizione-CNR ViaSanMartinodellaBattaglia44 00185Rome,Italy E-mail:{giovanni.pezzulo,cristiano.castelfranchi,rino.falcone}@istc.cnr.it MartinV.Butz UniversitätWürzburg InstitutfürPsychologie,KognitivePsychologieIII Röntgenring11,97070Würzburg,Germany E-mail:[email protected] LibraryofCongressControlNumber:2008934765 CRSubjectClassification(1998):I.2,I.6,F.1.2,G.3,K.8 LNCSSublibrary:SL7–ArtificialIntelligence ISSN 0302-9743 ISBN-10 3-540-87701-0SpringerBerlinHeidelbergNewYork ISBN-13 978-3-540-87701-1SpringerBerlinHeidelbergNewYork Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis concerned,specificallytherightsoftranslation,reprinting,re-useofillustrations,recitation,broadcasting, reproductiononmicrofilmsorinanyotherway,andstorageindatabanks.Duplicationofthispublication orpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawofSeptember9,1965, initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.Violationsareliable toprosecutionundertheGermanCopyrightLaw. SpringerisapartofSpringerScience+BusinessMedia springer.com ©Springer-VerlagBerlinHeidelberg2008 PrintedinGermany Typesetting:Camera-readybyauthor,dataconversionbyMarkusRichter,Heidelberg Printedonacid-freepaper SPIN:12531196 06/3180 543210 Foreword The general idea that brains anticipate the future, that they engage in prediction, andthatonemeansofdoingthisisthroughsomesortofinnermodelthatcanberun offline,hasalonghistory.SomeversionoftheideawascommontoAristotle,aswell as to many medievalscholastics, to Leibniz and Hume, and in more recenttimes, toKennethCraikandPhilipJohnson-Laird.Onereasonthatthisgeneralidearecurs continuallyisthatthisisthekindofpicturethatintrospectionpaints.Whenweare engagedintasksitseemsthatweformimagesthatarepredictions,oranticipations, andthattheseimagesareisomorphictowhattheyrepresent. But as much as the general idea recurs, opposition to it also recurs. The idea hasneverbeenwidelyaccepted,oruncontroversialamongpsychologists,cognitive scientistsandneuroscientists.Themainreasonhasbeenthatsciencecannotbesat- isfiedwithmetaphorsandintrospection.Inordertogainacceptance,anideaneeds tobeformulatedclearlyenoughsothatitcanbeusedtoconstructtestablehypothe- seswhoseresultswillclearlysupportorcastdoubtuponthehypothesis.Next,those ideasthatareformulableinoneoranothersortofsymbolismornotationarecapable ofbeingmodeled,andmodelingisahugepartofcognitiveneuroscience.Ifanidea cannotbeclearlymodeled,thentherearelimitstohowwidelyitcanbetestedand acceptedbyacognitiveneurosciencecommunity.Andfinally,ideally,theideawill be articulatedand modeledin such a way thatit is nota completemysteryhow it couldbeimplementedbythebrain.Thoughtheideathatthebrainmodelsandpre- dictsandanticipatesissupportedbyintrospectionandalonghistoryofhypotheses, ithaslargelyfailedontheselatterthreecounts–especiallycomparedwithvarious theoreticalcompetitors.And this is whythe extentto which ithas beenembraced bycognitivescienceandneurosciencehasbeenlimited. But there is good news. Mathematical tools from a number of areas, including moderncontroltheoryandsignalprocessing,arecapableofallowingforverypre- cisemathematicalformulationsofthebasicidea,aswellasmanyspecificversions. This allows for the ideas notonly to be precisely formulated,but also to be mod- eledandcomparedtohumanbehavioraldata.Andgivena numberofschemesfor implementingthesekindsofmathematicalmodelsinneuralsystems,itispossible VI Foreword toseethesemodelsasbeingimplementedinthebrain.Thequalitativeideathatthe brainmodelstheworldisfinallybeingclarifiedandquantified. But we are still in the early stages of this process. While there are many pro- posalsandtheoriesthatarebeginningtotakeshape,therehavebeenfewsustained treatmentsofthetopicthatattempttodevelopthemindetailedandconsistentways. Rather,theapplicationshavelargelybeenpiecemeal.Inthisregardthepresentvol- umerepresentsa significantadvanceinthe field.Itoffersa sustainedtreatmentof variousaspects of the generalhypothesis,notonlyin termsof being conceptually clearandconsistent,butalso intermsofpresentinga wide rangeof particularap- plicationsthatillustratetheconceptualmachineryinaction. Itwouldbeanoverstatementtosaythattheideathatthebrainisamodelerand predictorisrevolutionary,orthatthecurrentswellintheoreticalinterestintheidea representstheinitialstagesofarevolutionincognitiveneuroscience.Butwhiletalk ofrevolutionmaybeoverstatement,itcannotbedeniedthatthisapproachtounder- standingbrainfunctionisbeginningtotakeonanimportancecomparabletothatof traditionalartificialintelligenceapproachesandconnectionistmodelingapproaches. The clarity, detail and quality of the ideas presented in this volume, coupled with thegrowingimportanceofthisgeneralapproach,makethisvolumeacriticalcontri- butiontoourunderstandingofbrainfunction,andshouldbereadbyanyonewitha seriousinterestinunderstandinghowthebrainmanagestosupportcognitivefunc- tions. July2008 RickGrush UniversityofCalifornia,SanDiego Preface Predictionisdifficult–especiallyforthefuture.NielsBohr Over the last few decades, it has become increasingly clear that animals most of the time do not simply react in their world based on unconditioned or condi- tionedstimuli,butratheractivelyoperateintheirenvironmentinahighlygoal-and future-orientedway,andnotjustonthebasisofcurrentperception,butinpartau- tonomouslyfromenvironmentalstimuli.Psychologynowsuggeststhatitisthegoal itselfthattriggersbehaviorandattention.Learningishighlyinfluencedbycurrent predictive knowledgeand the consequentdetection of novelty.Behavioral control ismosteffectivelycontrolledbythehelpofforwardmodelsthatsubstitutedelayed orthatenhancenoisyperceptualfeedback.Thus,anticipationscomeinmanyforms andinfluencemanycognitivemechanisms. Thisbookproposesaunifyingapproachfortheanalysisanddesignofartificial cognitivesystems:Theanticipatoryapproach.Weproposeafoundationalviewof theimportanceofdealingwiththefuture,ofgainingsomeautonomyfromcurrent environmentaldata,ofendogenouslygeneratingsensorimotorandabstractrepresen- tations.Weproposeameaningfultaxonomyofanticipatorycognitivemechanisms, distinguishingbetweenthetypesofpredictionsandthedifferentinfluencesofthese predictionsonactualbehaviorandlearning.Doingso,wesketchoutanew,unify- ing perspective on cognitivesystems. Mechanisms, that have often been analyzed inisolationorhavebeenconsideredunrelatedtoeachother,nowfitintoacoherent wholeandcanbeanalyzedincorrelationtoeachother.Learningandbehaviorare considered increasingly intertwined and correlated with each other. Attention and action control suddenly appear as very similar processes. Goal-oriented behavior, motivationandemotionappearasrelatedandintertwined. While the revelation of these correlations is helpful for the analysis and com- parisonofdifferentlearningandbehavioralmechanisms,thesecondbenefitofthe anticipatoryapproachisthepossibilitytomodularlydesignnovelcognitivesystem architectures.Thedevelopedtaxonomyclearlycharacterizeswhichaspectsareim- portant for different anticipatory cognitive modules and how these modules may interact with each other. Thus, the second benefit of the anticipatory approach is VIII Preface thefacilitationofcognitivesystemdesign.Buildingblocksofcognitivesystemsare proposedandexemplarilyanalyzedindiversesystemarchitectures.Theinteraction ofthesebuildingblocksthenischaracterizedbytheiranticipatorynature,facilitat- ing the design of larger, more competentautonomousartificial cognitive systems. We hope that the proposed anticipatory approach may thus not only serve for the analysisofcognitivesystemsbutratheralsoasaninspirationandguidelineforthe progressivelymoreadvancedandcompetentdesignoflarge,butmodular,artificial cognitivesystems. Acknowledgments This work is supportedby the EU project MindRACES, from Reactive to Antic- ipatory Cognitive Embodied Systems, funded under grant FP6-511931 under the “CognitiveSystems” initiativefrom the EC. Specialthanksto ourProjectOfficer, Ce´cileHuet,andtoourProjectReviewers,LolaCan˜ameroandDeepakKumar,for theirvaluableencouragementandadvice. July2008 GiovanniPezzulo MartinV.Butz CristianoCastelfranchi RinoFalcone Contents PartI Theory 1 Introduction:AnticipationinNaturalandArtificialCognition...... 3 GiovanniPezzulo, Martin V. Butz, Cristiano Castelfranchi, and RinoFalcone 1.1 Introduction.............................................. 3 1.2 ThePathtoAnticipatoryCognitiveSystems................... 4 1.2.1 SymbolicBehavior,Representation-LessBehavior,and TheirMergetoAnticipatoryBehavior ................ 5 1.2.2 ThePowerofAnticipation:FromReactivitytoProactivity 6 1.2.3 TheAnticipatoryApproachtoCognitiveSystems....... 6 1.2.4 TheUnitaryNatureofAnticipation................... 12 1.3 AnticipationinLivingOrganisms............................ 12 1.3.1 AnticipatoryNaturalCognition...................... 12 1.3.2 AnticipatoryCodesintheBrain ..................... 15 1.3.3 SimulativeTheoriesofCognition,andTheirUnifying Nature........................................... 18 1.4 Conclusions.............................................. 22 2 TheAnticipatoryApproach:DefinitionsandTaxonomies .......... 23 GiovanniPezzulo,MartinV.Butz,andCristianoCastelfranchi 2.1 AnticipatorySystems,Anticipation,andAnticipatoryBehavior... 23 2.2 Predictionvs.Anticipation ................................. 25 2.2.1 PredictiveCapabilities ............................. 25 2.2.2 AnticipatoryCapabilities ........................... 31 2.3 AnticipationandGoal-OrientedBehavior ..................... 34 2.3.1 TheAnticipatoryStructureofGoal-OrientedBehavior .. 35 2.3.2 NotAllAnticipatoryBehaviorIsGoal-Oriented........ 36 2.3.3 WhichAnticipationsPermitGoal-OrientedAction?..... 36 2.3.4 The Hierarchical Organization of Anticipatory Goal-OrientedAction .............................. 37 X Contents 2.3.5 AdditionalElementsofTrueGoal-OrientedBehavior ... 38 2.4 AnticipationandLearning.................................. 39 2.4.1 LearningtoPredict ................................ 39 2.4.2 BootstrappingAutonomousCognitiveDevelopment: SurpriseandCuriosity ............................. 40 2.4.3 FromWilledtoAutomaticControlofActionandVice VersaontheBasisofSurprise ....................... 41 2.5 Conclusions.............................................. 43 3 BenefitsofAnticipationsinCognitiveAgents .................... 45 MartinV.ButzandGiovanniPezzulo 3.1 PotentialsforAnticipatorySystems .......................... 45 3.2 PotentialBenefitsofAnticipatoryMechanismsonCognitive Functions................................................ 48 3.2.1 Effective,Context-BasedActionInitiation............. 48 3.2.2 FasterandSmootherBehaviorExecution.............. 49 3.2.3 ImprovingTop-DownAttention ..................... 50 3.2.4 ImprovingInformationSeeking...................... 51 3.2.5 ImprovingDecisionMaking ........................ 52 3.2.6 ObjectGrounding,Categorization,andOntologies...... 54 3.2.7 SocialAbilities ................................... 55 3.2.8 Learning......................................... 57 3.3 ArisingChallengesDuetoAnticipationsandAvoidingThem .... 60 3.4 Conclusion............................................... 61 PartII Models,Architectures,andApplications 4 AnticipationinAttention ..................................... 65 Christian Balkenius, Alexander Fo¨rster, Birger Johansson, and VinThorsteinsdottir 4.1 Introduction.............................................. 65 4.2 LearningWhattoLookat .................................. 66 4.2.1 ALearningSaliencyMap........................... 67 4.3 Cue-TargetLearning....................................... 70 4.3.1 CueingbyaSingleStimulus ........................ 70 4.3.2 ContextualCueing................................. 72 4.3.3 FoveaBasedSolution .............................. 72 4.4 AttendingtoMovingTargets................................ 73 4.4.1 ModelsofSmoothPursuit .......................... 75 4.4.2 EngineeringApproaches............................ 76 4.4.3 TheStateBasedApproach.......................... 78 4.4.4 ThePredictionApproach ........................... 79 4.4.5 TheFoveaBasedApproach ......................... 80 4.5 CombiningBottom-UpandTop-DownProcesses............... 81 Contents XI 5 Anticipatory,Goal-DirectedBehavior .......................... 85 MartinV.Butz,OliverHerbort,andGiovanniPezzulo 5.1 ABriefHistoryofSchemas................................. 87 5.2 SchemaApproaches....................................... 88 5.2.1 SymbolicSchemasforPolicyLearning ............... 89 5.2.2 SymbolicSchemasandPredictionforSelection ........ 90 5.2.3 Neural-BasedPlanning............................. 91 5.2.4 NeuralNetwork-BasedDynamicProgramming ........ 92 5.3 InverseModelApproaches ................................. 92 5.3.1 InverseModelsinMotorLearningandControl......... 93 5.3.2 InverseModelsandSchemaApproaches .............. 94 5.4 AdvancedStructures....................................... 94 5.4.1 PredictionandAction .............................. 95 5.4.2 CoupledForward-InverseModels .................... 97 5.4.3 HierarchicalAnticipatorySystems ................... 98 5.5 EvaluationofPredictiveandAnticipatoryCapabilities .......... 99 5.5.1 Schema-BasedSystems ............................101 5.5.2 InverseModelApproaches..........................106 5.6 Discussion ...............................................108 5.6.1 ContrastingPredictiveSystemCapabilities ............108 5.6.2 ContrastingAnticipatorySystemCapabilities ..........110 5.6.3 Integration .......................................112 5.7 Conclusions..............................................113 6 AnticipationandBelievability ................................. 115 CarlosMartinhoandAnaPaiva 6.1 Introduction..............................................115 6.1.1 AnimationandBelievability ........................115 6.1.2 EmotionandExaggeration..........................116 6.1.3 Anticipation......................................117 6.1.4 Anticipation,Emotion,andBelievability ..............117 6.2 RelatedWork.............................................119 6.2.1 OzProject........................................119 6.2.2 EMA............................................119 6.2.3 DuncantheHighlandTerrier ........................120 6.3 Emotivector..............................................121 6.3.1 Architecture......................................121 6.3.2 AnticipationModel................................122 6.3.3 SalienceModel ...................................123 6.3.4 SensationModel ..................................123 6.3.5 SelectionModel...................................123 6.3.6 Uncertainty.......................................124 6.4 Aini,theSyntheticFlower..................................125 6.4.1 EmotivectorsinAction.............................125 6.4.2 Evaluation .......................................128

Description:
This book proposes a unifying approach for the analysis and design of artificial cognitive systems: The Anticipatory Approach. In 11 coherent chapters, the authors of this State-of-the-Art Survey propose a foundational view of the importance of dealing with the future, of gaining some autonomy from
See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.